Edinburgh Research Explorer · changes in the left and right ATLs, there was increased effec-tive connectivity from the right ATL to the left ATL (Jung and Lambon Ralph 2016). In

Edinburgh Research Explorer

Revealing the dynamic modulations that underpin a resilientneural network for semantic cognition

Citation for published version:Rice, GE, Caswell, H, Moore, P, Lambon Ralph, MA & Hoffman, P 2018, 'Revealing the dynamicmodulations that underpin a resilient neural network for semantic cognition: An fMRI investigation in patientswith anterior temporal lobe resection', Cerebral Cortex, vol. 28, no. 8, bhy116, pp. 3004-3016.https://doi.org/10.1093/cercor/bhy116

Digital Object Identifier (DOI):10.1093/cercor/bhy116

Link:Link to publication record in Edinburgh Research Explorer

Document Version:Publisher's PDF, also known as Version of record

Published In:Cerebral Cortex

General rightsCopyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s)and / or other copyright owners and it is a condition of accessing these publications that users recognise andabide by the legal requirements associated with these rights.

Take down policyThe University of Edinburgh has made every reasonable effort to ensure that Edinburgh Research Explorercontent complies with UK legislation. If you believe that the public display of this file breaches copyright pleasecontact [email protected] providing details, and we will remove access to the work immediately andinvestigate your claim.

Download date: 21. Feb. 2020

https://www.research.ed.ac.uk/portal/en/persons/paul-hoffman(21d7d437-f5e9-49e9-b0dc-cdc9c74d4e02).html

https://www.research.ed.ac.uk/portal/en/publications/revealing-the-dynamic-modulations-that-underpin-a-resilient-neural-network-for-semantic-cognition(c4d1441a-a9a1-4eb3-8e14-ae3aee67c2fd).html



https://doi.org/10.1093/cercor/bhy116

https://doi.org/10.1093/cercor/bhy116


Cerebral Cortex, 2018; 1–13

doi: 10.1093/cercor/bhy116Original Article

O R I G I NA L ART I C L E

Revealing the Dynamic Modulations That Underpin aResilient Neural Network for Semantic Cognition: AnfMRI Investigation in Patients With Anterior TemporalLobe ResectionGrace E. Rice 1, Helen Caswell2, Perry Moore3,Matthew A. Lambon Ralph 1 and Paul Hoffman 4

1Neuroscience and Aphasia Research Unit (NARU), University of Manchester, Manchester M13 9PL, UK,2Department of Clinical Neuropsychology, Salford Royal Hospital, Manchester M6 8HD, UK, 3Department ofClinical Neuropsychology, The Walton Centre NHS Foundation Trust, Liverpool L9 7LJ, UK and 4Centre forCognitive Ageing and Cognitive Epidemiology (CCACE), Department of Psychology, University of Edinburgh,Edinburgh EH8 9JZ, UK

Address correspondence to Dr Grace Rice, Prof. Matthew A. Lambon Ralph, or Dr Paul Hoffman, Neuroscience & Aphasia Research Unit (NARU), Divisionof Neuroscience and Experimental Psychology, School of Biological Sciences, University of Manchester, Brunswick Street, Manchester M13 9PL, UK.Email: [email protected] orcid.org/0000-0002-0847-9066, orcid.org/0000-0001-5907-2488, orcid.org/0000-0002-3248-3225

AbstractOne critical feature of any well-engineered system is its resilience to perturbation and minor damage. The purpose of thecurrent study was to investigate how resilience is achieved in higher cognitive systems, which we explored through thedomain of semantic cognition. Convergent evidence implicates the bilateral anterior temporal lobes (ATLs) as a conceptualknowledge hub. While bilateral damage to this region produces profound semantic impairment, unilateral atrophy/resectionor transient perturbation has a limited effect. Two neural mechanisms might underpin this resilience to unilateral ATLdamage: 1) the undamaged ATL upregulates its activation in order to compensate; and/or 2) prefrontal regions involved incontrol of semantic retrieval upregulate to compensate for the impoverished semantic representations that follow from ATLdamage. To test these possibilities, 34 postsurgical temporal lobe epilepsy patients and 20 age-matched controls werescanned whilst completing semantic tasks. Pictorial tasks, which produced bilateral frontal and temporal activation, showedfew activation differences between patients and control participants. Written word tasks, however, produced a left-lateralized activation pattern and greater differences between the groups. Patients with right ATL resection increasedactivation in left inferior frontal gyrus (IFG). Patients with left ATL resection upregulated both the right ATL and right IFG.Consistent with recent computational models, these results indicate that 1) written word semantic processing in patientswith ATL resection is supported by upregulation of semantic knowledge and control regions, principally in the undamagedhemisphere, and 2) pictorial semantic processing is less affected, presumably because it draws on a more bilateral network.

Key words: anterior temporal lobectomy, conceptual knowledge, laterality, semantic memory, temporal lobe epilepsy

© The Author(s) 2018. Published by Oxford University Press.This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/),which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.

Downloaded from https://academic.oup.com/cercor/advance-article-abstract/doi/10.1093/cercor/bhy116/5033554by The University of Edinburgh useron 08 June 2018

http://www.oxfordjournals.org

http://orcid.org/0000-0002-0847-9066

http://orcid.org/0000-0001-5907-2488

http://orcid.org/0000-0002-3248-3225

http://orcid.org/0000-0002-0847-9066

http://orcid.org/0000-0002-0847-9066

http://orcid.org/0000-0001-5907-2488

http://orcid.org/0000-0001-5907-2488

http://orcid.org/0000-0002-3248-3225

http://orcid.org/0000-0002-3248-3225

http://creativecommons.org/licenses/by/4.0/

IntroductionOne critical feature of any well-engineered system is its resil-ience to perturbation and minor damage. How resilience isachieved in higher cognitive systems is important both for cog-nitive and clinical neuroscience. Although rarely considered inlaboratory-based explorations of higher cognition, in everydaylife we are faced with and are resilient to variations in task dif-ficulty, degraded stimuli, etc. Likewise after partial brain dam-age or perturbation, patients can sometimes show impressiveresilience and recovery. The current study examined thedynamic changes that support resilience in the domain ofsemantic cognition—both in patients with partial damage tothe semantic representational system and in the healthy sys-tem following variation in task difficulty.

Unilateral resection is the primary surgical treatment fordrug-resistant temporal lobe epilepsy (TLE) (Wiebe et al. 2001).Surgical resection alleviates seizure activity in 80–90% ofpatients making it the gold standard of treatment (Wiebe et al.2001). Behaviorally, patients with unilateral anterior temporallobe (ATL) resection exhibit short-term memory problems andrelatively mild semantic memory impairments (Wilkins andMoscovitch 1978; Alpherts et al. 2006; Lambon Ralph et al. 2012;Willment and Golby 2013; Sidhu et al. 2015). This mild semanticimpairment is shown through slower response times onsemantic tasks (e.g., picture naming, word-picture matching,synonym judgment), rather than marked decreases in accuracy(Wilkins and Moscovitch 1978; Lambon Ralph et al. 2012; Riceet al. 2018), although on more challenging/specific level seman-tic concepts decreases in accuracy performance are found. Thismild semantic impairment in patients with unilateral ATLresection is in contrast to the severe semantic impairmentshown after bilateral ATL damage/resection (Terzian and Ore1955; Lambon Ralph et al. 2010, 2012; Mion et al. 2010; Zhaoet al. 2017). This pattern of results was also shown in seminalnonhuman primate work (Brown and Schafer 1888; Kluver andBucy 1937, 1938), in which unilateral resection resulted in tran-sient deficits, whereas bilateral resection resulted in severemultimodal semantic and episodic memory impairments.Converging evidence from functional neuroimaging and neuro-stimulation studies indicates that the bilateral ATLs play a criti-cal and central role in semantic representation (Pobric et al.2010; Visser, Embleton et al. 2010; Shimotake et al. 2014; Rice,Lambon Ralph, and Hoffman 2015; Lambon Ralph et al. 2017).The discrepancy in behavioral performance in patients withunilateral versus bilateral ATL damage/atrophy imply that thissystem is configured to be robust and relatively resistant tounilateral damage (Schapiro et al. 2013). Despite this, the neu-ronal mechanisms underlying this resilience are not clear. Twononexclusive and potentially related explanations for therobustness of the semantic system are: 1) after unilateral resec-tion the remaining contralateral ATL changes its activation inorder to compensate; and/or 2) performance on semantic tasksis maintained through changes in the activation in otherregions of the semantic network.

The first potential explanation for the robustness of thesemantic system is that the system has a certain amount ofredundancy when semantic representations are distributed acrossboth left and right ATLs. As such it would follow that, in patientswith unilateral damage/resection, the undamaged contralateralATL would still retain considerable representational effectiveness.As a consequence, the semantic impairment in TLE is primarilyshown through slower response times, rather than as markeddecreases in accuracy (Lambon Ralph et al. 2010; Schapiro et al.

2013). Formal computational explorations of a bilaterally config-ured ATL system not only showed the same behavioral differencebetween unilateral (limited deficits) versus bilateral damage (con-siderable impairment) even when total damage is matched, butalso found a second important factor (Schapiro et al. 2013).Specifically, damage to a local neural network not only weakenedthe distributed representation but it also resulted in activationnoise which can be propagated to connected units. Accordingly, inunilateral lesions, noise was propagated to other ipsilateral hubunits but less was transmitted to the distant contralateral hub.After bilateral damage, noise was pervasive throughout the repre-sentational system.

Evidence for a critical role of the contralateral hemisphereafter unilateral perturbation has been observed in concurrentfMRI-TMS studies in healthy participants (Binney and LambonRalph 2015; Jung and Lambon Ralph 2016). Repetitive transcra-nial magnetic stimulation to the left ATL diminished activationin this region during semantic processing (mimicking unilateralATL damage) and increased activation in the nonstimulatedright ATL. Jung and Lambon Ralph (2016) also employed effec-tive connectivity analysis and found that it is not just theincreased activation of the contralateral ATL that is importantin maintaining normal semantic performance, but the connec-tivity between the left and right ATLs. Specifically, after leftATL theta-burst stimulation, in addition to the activationchanges in the left and right ATLs, there was increased effec-tive connectivity from the right ATL to the left ATL (Jung andLambon Ralph 2016). In a combined fMRI-neuropsychologicalstudy, Warren et al. (2009) also highlighted the importance ofthe connectivity between the ATLs in supporting language per-formance. In a group of chronic poststroke aphasic patients,these authors observed an overall decrease in inter-ATL con-nectivity. However, patients who exhibited greater preservationof this connectivity showed better language/comprehensionperformance.

A second type of explanation for the robustness of semanticperformance is that there is increased activation of the regionsinvolved in control and selection of semantic knowledge, therebycompensating for the reduced quality of semantic representa-tions that follow from ATL dysfunction/damage. The inferiorfrontal gyrus (IFG) shows strong responses to increased semanticprocessing demand and has been implicated in the executive-control component of semantic cognition (Thompson-Schill et al.1997; Badre and Wagner 2002; Noonan et al. 2013). In the contextof patients with TLE, regions in the bilateral IFG were found to beupregulated in semantic tasks both before surgery (Billingsleyet al. 2001; Koylu et al. 2006; Bonelli et al. 2012; Rosazza et al.2013) and after ATL resection surgery (Backes et al. 2005;Noppeney et al. 2005; Koylu et al. 2006; Wong et al. 2009; Bonelliet al. 2012; Tivarus et al. 2012). This increase of IFG activation isnot uniform; rather, activation in the IFG interacts with the site ofresection. Patients who have undergone right ATL resection acti-vate similar left IFG regions to control participants, but do somore strongly. In contrast, patients who have undergone left ATLresection show reduced activation in the left IFG during expres-sive semantic tasks and increased activation in the right IFG com-pared with controls (Backes et al. 2005; Koylu et al. 2006; Wonget al. 2009; Bonelli et al. 2012). This shift in IFG activation follow-ing left ATL resection has been shown to be behaviorally relevant,such that stronger activation in the right IFG is correlated withbetter naming performance after surgery (Bonelli et al. 2012).

In addition, upregulation of the IFG shows dynamic changesdepending on the modality of the task. Episodic encoding ofwritten words in preoperative left TLE patients has been

2 | Cerebral Cortex


associated with bilateral IFG activation, in comparison to theleft-lateralized pattern of response in preoperative right TLEpatients and control participants (Maccotta et al. 2007).Contrastively, encoding of nonverbal material generated rightlateralized frontal responses in both left and right TLE patientsand control participants (Maccotta et al. 2007). This study alsoshowed that after surgery, the left TLE patients regained moreof a left-lateralized activation pattern in the verbal encodingtask (however, this neural change did not have a correspondingbehavioral change in terms of accuracy on the task). Thisimplies that changes in prefrontal activation after unilateralresection may interact with the task being performed, as wellas the laterality of resection.

Within the IFG, there are regions hypothesized to underliethe cognitive control of language specifically versus othersinvolved in domain-general cognitive control (Badre et al. 2005;Badre and D’Esposito 2009; Duncan 2010). This raises the possi-bility that recovery after unilateral resection may not be under-pinned by semantic/language specific processes but by theupregulation of domain-general cognitive resources. Evidencefor this in the context of stroke aphasia comes from Brownsettet al. (2014), who showed that during an auditory comprehen-sion task stroke aphasic patients upregulated regions in themidline frontal cortex (encompassing the dorsal anterior cingu-late cortex and superior frontal gyrus). This region forms partof the domain-general cognitive control network and was neu-roanatomically distinct from the language-specific IFG activa-tion in control participants when performing the same task. Anextension of this line of thinking suggests that the domain-general cognitive control regions that are upregulated afterdamage are also the same regions healthy participants use dur-ing more challenging processing conditions (Sharp et al. 2004;Brownsett et al. 2014; Geranmayeh et al. 2014). In other words,changes in the system after damage are not random but mayreflect intrinsic mechanisms present in the healthy brain. Totest this hypothesis, Brownsett et al. (2014) tested control parti-cipants on a more challenging auditory comprehension taskusing noise-vocoded speech. Under these more demandingconditions, controls performed less accurately and, critically,showed increased activation in the same midline frontalregions that were upregulated in the stroke patients. Brownsettet al. (2014) concluded that recovery of language processingafter stroke was driven by domain-general cognitive controlregions, which are also recruited by healthy individuals in thecontext of increased processing demands. It is possible thatsimilar mechanisms are engaged by resected TLE patients tocompensate for the impaired function of the ATL semanticsystem.

This study was designed to investigate whether one or bothof these resilience mechanisms (shift in the division of seman-tic labor across the ATLs; and/or engagement of executive con-trol) is found in patients with resection for TLE. The specificgoals and features of the study were as follows: 1) determinewhether the increased activity in prefrontal regions reflectssubregions associated with semantic executive processes ordomain-general cognitive control. 2) Explore the relativeengagement of ipsilateral and contralateral ATL/temporalregions We note that, in the small handful of previous fMRIexplorations of patients with resection for TLE, the focus hasprimarily been on speech production (naming, verbal fluency,etc.) and on the changes observed in frontal regions. There is apaucity of data on the primary region of interest for TLE,namely, the ATL, though explorations using combined rTMS-fMRI suggest there might be important changes in this region

(Schapiro et al. 2013; Binney and Lambon Ralph 2015; Jung andLambon Ralph 2016). This lack of data might be related to a setof methodological challenges associated with successful fMRIinvestigations of the ATL including fMRI signal drop-out anddistortion, the use of active baselines and ensuring a full field-of-view (Devlin et al. 2000; Visser, Jefferies, and Lambon Ralph2010). Accordingly, all of these methodological issues wereaddressed in the current investigation, including the use ofdual-echo EPI imaging to improve signal in the ATLs (Halaiet al. 2014). 3) As noted above, previous explorations of episodicmemory indicate that the fMRI changes are dynamic—changingaccording to modality of presentation. This has never beenexplored in the context of semantic function. Given the recentcomputational explorations of bilateral versus unilateral ATL-hub function, we predicted that a bilaterally distributed task(i.e., pictorial semantic processing) may be more robust to uni-lateral resection, compared with a relatively left-lateralizedtask (e.g., written word semantic processing).

These study goals were met by exploring whether changes inactivation interacted with 1) side of resection (left vs. right ATLresection) and 2) the modality of the task (written words vs. pic-tures). Semantically related activation in the contralateral ATLand other regions in the semantic network were comparedbetween resected TLE patients and a group of age-matched con-trol participants. To explore whether control participants alsoupregulated a similar network of regions under increased taskdemands, a difficulty manipulation was employed in the controlgroup. One important point to note is that the current studyinvolved only postsurgical data from resected TLE patients. Thismeans that there are a number of potential explanations for anydifference found between patients and healthy controls, includ-ing the resection surgery itself and to the presence of long-standing epilepsy affecting functional networks prior to surgery.We consider these possibilities in more detail in the Discussion.

MethodPatients

A total of 34 patients who had a single “en-bloc” unilateralresection for medically intractable epilepsy took part in thestudy (17 left TLE, 17 right TLE). All patients underwent exten-sive neuropsychological testing (Table 1). Briefly, both the leftand right TLE patients were matched on age (t[32] = 0.56, P =0.58), education (t[32] = 0.80, P = 0.43), epilepsy duration (t[32] =0.28, P = 0.78), age at surgery (t[32] = 0.27, P = 0.79), age at diag-nosis (t[32] = 0.02, P = 0.98), and number of antiepileptic drugs(AEDs: t[32] = 1.00, P = 0.32). All patients were in the chronicstage postsurgery (at least 1 year postsurgery), although theright TLE patients were slightly more chronic than the left TLEgroup (t[32] = 2.11, P = 0.04). All patients had late-onset epi-lepsy, no history of other neurological or psychiatric disorders,were left language dominant based on the results of the WADAtest and had epilepsy arising from unilateral mesial temporalsclerosis. All were native English speakers, right handed withnormal or corrected-to-normal vision. The study was approvedby the local ethics board.

Control Participants

The pattern of brain activation and behavioral performance inthe scanner of the left and right TLE patients were comparedseparately to a group of 20 age-matched control participants(controls versus left TLE: t[35] = 1.19, P = 0.24; controls versusright TLE: t[35] = 1.85, P = 0.07). The TLE patients had completed

Revealing the Dynamic Modulations That Underpin a Resilient Neural Network for Semantic Cognition Rice et al. | 3


marginally less formal education than controls, consistent withtheir long-standing neurological condition (controls versus leftTLE: t[35] = 3.01, P = 0.005; controls vs. right TLE: t[35] = 1.69, P =0.10). All control participants were right-handed, native Englishspeakers and who had normal or corrected-to-normal vision.Control participants also underwent extensive neuropsycholog-ical testing to screen for any undiagnosed cognitive abnormali-ties. The experiment was approved by the local ethics board.

Stimuli and Task

Participants underwent 2 tasks in the scanner: a semantic asso-ciation test and a test of occupation matching in famous people(Fig. 1). The semantic association task was the Camel andCactus test (Bozeat et al. 2000; Visser et al. 2012). On each trialparticipants were presented with a probe item and asked todecide which of 2 alternatives was semantically related. Thesecond task was an occupation judgment task. On each trialparticipants were presented with a probe item and asked todecide which of 2 alternatives had the same occupation. Foreach trial on the occupation matching task, all items sharedthe same gender. To investigate potential laterality differencesdue to modality, both the semantic association task and theoccupation matching task were presented as written words andpictures. The written word condition was included to reflect a“left lateralized” semantic task based on the previous literature(Gainotti 2012; Rice, Hoffman, and Lambon Ralph 2015; Rice,Lambon Ralph et al. 2015), pictures of famous faces wereincluded to reflect a “right lateralized” semantic task, again

based on the predictions from the previous literature (Gainotti2007; 2012). However, we failed to see any differences betweenpictures of objects and famous faces (see Supplementary FigsS1 and S2), therefore these 2 conditions were collapsed to forma single “picture” condition. Different items were used in theword and picture versions to avoid priming effects. Each condi-tion contained 33 items.

In addition to the semantic conditions, a visual matchingtask was used to account for low level visual effects and pro-vide an attention-demanding baseline condition. Baselineitems were generated by visually scrambling items from eachof the 4 semantic conditions; these were created using the JavaRuntime Environment (www.SunMicrosystems.com) by scram-bling each image into 120 pieces and rearranging them in a ran-dom order. For the written word trials (CCw, famous names)the baseline condition was scrambled words and for the picturetrials (CCp, famous faces) the baseline condition was scrambledversions of the picture stimuli. The baseline task was presentedin the same way as the experimental condition, with 1 probeitem and 2 choices. One of the choices was identical to the tar-get item and the other was an inverted version of the target.Participants were asked to decide which of the 2 choices wasidentical to the probe item.

Procedure

TLE patients and controls underwent 4 functional scans, eachwith a total scan time of 8.45min. During scanning, stimuliwere presented in a block design using E-Prime software

Table 1 Demographic information and background behavioral testing for the control participants and TLE patient groups. Values areexpressed as mean (standard deviation)

Controlsn = 20

Left TLEn = 17

Right TLEn = 17

Age (years) 38.2 (12.2) 42.9 (11.6) 44.9 (9.2)Education (years) 17.1 (2.2) 14.8 (2.5) 15.6 (3.4)Gender (M:F) 11:9 9:8 9:8Age at surgery (years) – 38.3 (11.2) 37.3 (10.1)Years since surgery – 4.6 (4.1) 7.5 (3.9)Age at diagnosis (years) – 15.7 (8.0) 15.6 (8.6)Epilepsy duration (years) – 22.6 (11.2) 21.6 (9.9)Number of AEDs – 2.5 (1.2) 1.8 (1.3)Volume resected (mm3) – 39.3 (8.7) 64.9 (22.3)Background behavioral testingWASI

Overall – 91.9 (12.1) 101.4 (14.9)Verbal – 84.0 (18.2) 98.2 (8.4)Matrix Reasoning – 100.3 (13.7) 104.1 (11.4)

Rey complex figureCopy (max = 36) 35 (3) 35 (1) 35 (1)Immediate (max = 36) 21 (8) 17 (7)* 14 (8)*Delayed (max = 36) 22 (9) 16 (7)* 14 (8)*

Digit spanForward (max = 9) 6.8 (0.9) 6.1 (1.1) 6.8 (1.1)Backward (max = 9) 4.6 (1.1) 4.4 (0.9) 4.2 (1.1)

Camden episodic memoryWords (max = 25) 23 (2) 20 (3)* 22 (3)Faces (max = 25) 23 (1) 21 (5) 21 (3)Picture naming (max = 38) 34 (6) 26 (6)*,** 30 (4)*WPM (max = 46) 43 (2) 39 (4)* 41 (3)Synonym judgment (max = 96) 89 (6) 79 (6)*,** 87(6)

*P < 0.05 versus controls; **P < 0.05 versus TLE group.

4 | Cerebral Cortex


http://www.SunMicrosystems.com

(Psychology Software Tools). Each functional scan contained sti-muli from one semantic condition (CCw, CCp, famous names orfamous faces) and from the relevant baseline condition (scram-bled pictures or scrambled words). This was done to avoid taskswitching effects in the scanner. Each block contained 3 trialsfrom 1 experimental condition. Each stimulus and the responsescreen were presented for 5000ms, with an interstimulus inter-val of 500ms. The 2 experimental conditions (semantic andbaseline) were sampled 11 times per functional scan in a coun-terbalanced order, giving a total of 22 blocks per scan. The orderof the scans was randomized and counterbalanced across parti-cipants. Stimuli were presented visually via a mirror mountedon the head coil, angled at a screen at the foot of the scannerbed. All participants underwent practice trials before beginningthe scan to familiarize them with the tasks.

To test whether any upregulation in activation in the TLEpatients were in regions that respond to increased task demands,we manipulated task difficulty for the control participants. Inaddition to the 4 functional scans in which stimuli were pre-sented at the same rate as for the patients (Controls [Matched]),control participants also completed an additional 4 functionalscans at a faster presentation speed (Controls [Speeded]). Thesame number of trials was presented as described above, but diffi-culty was increased by presenting stimuli twice as quickly, at2500ms intervals. Different stimuli were used in the slower andfaster versions of the task by switching modality of presentation.For example, trials which were presented as pictures in the slowscan were presented as words in the faster scan. The slower andfaster functional scans in the control participants were inter-leaved to avoid any habituation to the speed of presentation.Most control participants reported that the faster speed was morechallenging but was not so difficult as to prevent them from com-pleting the task (see also Behavioral Results).

Due to a technical issue with the button box in the scanner,accuracy data could only be recorded in 16 out of 17 left TLEpatients and 19 out of 20 control participants. Similarly, reac-tion time data could only be recorded in 15 out of 17 left TLEpatients and 19 out of 20 control participants (see BehavioralResults for full details).

Imaging Parameters

Traditionally, imaging the ventral ATLs has been problematicbecause of a number of technical issues including the nature of

the baseline contrast tasks as well as gradient-echo EPI signaldrop-out and distortion (Devlin et al. 2000; Visser, Jefferies et al.2010). These issues have been tackled through recent methodo-logical developments (Embleton et al. 2010; Visser, Embletonet al. 2010). In the current study, the core semantic task wascontrasted against an active baseline using dual-echo EPI imag-ing to improve signal in the vATLs (Halai et al. 2014).

All scans were acquired on a 3 T Phillips Achieva scanner,with a 32-channel head coil with a SENSE factor of 2.5. A dual-echo EPI sequence was used to improve signal-to-noise (SNR)in the vATLs (Halai et al. 2014). Using this technique, each scanconsisted of 2 images acquired simultaneously with differentecho times: a short echo optimized to obtain signal from thevATLs and a long echo optimized for good whole-brain cover-age. The sequence included 31 slices covering the whole brainwith repetition time (TR) = 2.8 s, echo times (TE) = 12 and 35ms,flip angle = 85°, FOV = 240 × 240mm2, resolution matrix = 80 × 80,slice thickness = 4mm, voxel size = 3 × 3 × 4mm3. All functionalscans were acquired using a tilt, up to 45° off the AC-PC line, toreduce ghosting artefacts in the temporal lobes. For the TLEpatients, functional scans were collected in four 8.45min runs;each run acquired 177 dynamic scans (including 2 dummy scans,which were excluded). For the control participants, an additionalfour 4.3min runs were included, each acquiring 88 dynamic scans(including 2 dummy scans, which were excluded). To addressfield-inhomogenities, a B0 field-map was acquired using identicalparameters to the functional scans except for the following: TR =599ms, TEs = 5.19 and 6.65ms. A high resolution T1 weightedstructural scan was acquired for spatial normalization, including260 slices covering the whole brain with TR = 8.4ms, TE =3.9ms, flip angle = 8°, FOV = 240 × 191mm, resolution matrix =256 × 206, voxel size = 0.9 × 1.7 × 0.9mm3.

fMRI Analysis

Analysis was carried out using SPM8 (Wellcome Department ofImaging Neuroscience, London; www.fil.ion.ucl.ac.uk/spm).

Automated Lesion Identification Procedure

Automated outlines of the resection area were generated usingSeghier et al. (2008) modified segmentation-normalization pro-cedure, which is designed for use with brain-injured patientsand which identifies areas of lesioned tissue. Data from boththe TLE patients and the control participants were subjected tothe automated lesion identification procedure. Segmentedimages were smoothed with an 8mm full-width half maximumGaussian kernel as recommended by Seghier et al. (2008) andsubmitted to the automated lesion identification and definitionmodules using the default parameters. The automated methodinvolves initial segmentation and normalizing into tissue clas-ses of grey matter, white matter, CSF and an extra tissue classwhich codes for the presence of the resection area. Aftersmoothing, voxels that emerge as outliers relative to normalparticipants are identified and the union of these outliers pro-vides the “fuzzy lesion map,” from which the resection outlineis derived. The generated images were used to create the resec-tion overlap map in Figure 2. For our patient sample, in order toensure that the algorithm correctly identified the resection areaas an extra class of tissue (rather than as CSF); the procedurewas run twice for the TLE patients. The first iteration was runusing the default settings in the toolbox; on the second itera-tion the default mask was changed to correspond to the outputfrom the first iteration. This constrained the algorithm onto the

Figure 1. Examples of stimuli from the 4 semantic conditions. Stimuli were

presented as pictures or written words.



http://www.fil.ion.ucl.ac.uk/spm

resection area and allowed a more precise segmentation of theresection area. Overall, the right TLE patients resection volumewas larger than that in the left TLE patients (Table 1; t[32] =4.42, P < 0.0001). This is in keeping with the current surgicalstandards whereby resections to the left hemisphere are moreconservative to avoid disruption to the language centers (Wiebeet al. 2001).

Preprocessing

The functional images from the short and long echoes wereaveraged using custom MATLAB code (Halai et al. 2014). Thesecombined images were realigned (for individual subject motioncorrection) and unwarped (for field-map correction). The meanfunctional image was then coregistered to the individual parti-cipant’s T1 structural scan. The data were then spatially nor-malized using the deformation fields generated in the Seghiermethod (see above) to MNI space and smoothed using an 8mmGaussian FWHM kernel. The normalized data were thenmasked to only include grey matter voxels (excluding those inthe white matter and cerebral–spinal fluid); this grey mattermask was included as an explicit mask during the first levelanalysis. By default, SPM also employs an implicit mask whichremoves voxels which respond in the lowest 20% with respectto the mean signal across the brain. This step removes regionswith low signal due to magnetic susceptibility and signal dis-tortion. However, given our a priori hypotheses about the ATLs,a region with high magnetic susceptibility, and the fact thatdual-echo acquisition was employed to reduce signal loss, thisdefault threshold was removed so that all voxels would beincluded in the analysis (Halai et al. 2014).

At the individual subject level, contrasts of interest weremodeled using a box-car function convolved with the canonicalhemodynamic response function. Low frequency drifts wereremoved using a high-pass filter of 128 s. In each run, 2 sepa-rate regressors were modeled: 1) semantic condition (either:CCp, CCw, Famous Faces, Famous Names) and 2) baseline con-dition (either scrambled pictures or scrambled words).

Whole Brain Analysis

The semantic and baseline regressors described above were usedto create separate contrasts for the picture trials (pictures >scrambled pictures) by averaging across the CCp and famous

face functional runs. The same method was followed to create acontrast for the written word trials (written word > scrambledwords) by averaging across the CCw and famous name func-tional runs. For the whole brain analysis, individual contrastmaps were entered into a second-level random effects analysisusing one sample t-tests. This whole-brain map was thresholdedat P < 0.001 at the voxel level, with a FWE-corrected clusterthreshold of P < 0.05.

ROI Analyses

Given the specific hypothesis regarding upregulation in thecontralateral hemisphere after unilateral ATL resection, an apriori ROI analysis was employed to assess activation in theleft and right vATLs. Peak coordinates were taken from a sepa-rate study on semantic processing that employed a writtenword task (Binney et al. 2010) (MNI: −15 −30; 36 −15 −30). A rightvATL peak was created using the homologous coordinates. Thecomparisons of interest were between activation in the leftvATL in the right TLE group versus Controls (Matched) and inthe right vATL in the left TLE group versus Controls (Matched).Activations for the manipulation of difficulty in control partici-pants (Controls [Matched], Controls [Speeded]) were also com-pared in the left and right vATLs separately. To explore thepossibility of upregulation in cognitive (semantic) controlregions after unilateral ATL resection, 2 additional ROIs werecreated using peaks from previous studies of semantic andcognitive control. The first peak was centered in the pars trian-gularis portion of the IFG (BA45) and was taken from a meta-analysis of semantic control. This region responded more tohard > easy semantic trials (MNI: −45 19 18; 47 23 26) (Noonanet al. 2013). The second peak was centered on the pars orbitalisportion of the IFG (BA47) and was taken from a study of seman-tic control; this region has been associated with controlledretrieval of semantic knowledge (MNI: −45 27 −15; 45 27 −15)(Badre et al. 2005).

ResultsBehavioral Results

In the scanner, participants made semantic association judg-ments for the 4 semantic conditions and similarity judgmentsfor the perceptual baseline conditions (Table 2). In the written

Figure 2. Resection overlap map for the 17 left and 17 right TLE patients. Overlap of the resection areas defined by the Seghier et al. (Seghier et al. 2008) method. Left

TLE patients overlap is shown on the right of the image, right TLE patients overlap is shown on the left of the image. Color bars indicate the number of patients with

resection in that area. Warmer colors = greater overlap, cooler colors = less overlap.

6 | Cerebral Cortex


word domain, comparisons between the left TLE patients andthe Controls (Matched) condition revealed that the left TLEpatients were significantly less accurate on the written wordsemantic task compared with controls (t[33] = 3.43, P = 0.002).There were no group differences on the nonsemantic baselinetask (t[33] = 1.01, P = 0.32). The left TLE subgroup were slowercompared with controls on both tasks (semantic: t[32] = 5.77,P < 0.0001; control: t[32] = 2.58, P = 0.02). Comparisons betweenthe right TLE patients and the Controls (Matched) group forwritten words showed no group differences in terms of accu-racy (semantic: t[34] = 1.85, P = 0.07; baseline: t[34] = 0.75, P =0.46). As with the left TLE subgroup, the right TLE subgroupwere slower compared with controls across both tasks (seman-tic: t[34] = 6.36, P < 0.0001; baseline: t[34] = 5.80, P < 0.0001).Comparisons within the left and right TLE subgroups revealedno significant differences in terms of accuracy (semantic: t[31]= 1.66, P = 0.11; baseline: t[31] = 1.70, P = 0.10) or reaction time(semantic: t[30] = 0.07, P = 0.95; baseline: t[30] = 1.82, P = 0.08)indicating that both groups were similarly impaired for thewritten word trials.

In the picture trials, comparisons between the left TLEpatients and Controls (Matched) revealed no differences in accu-racy between the 2 groups for the semantic task (t[33] = 1.58, P =0.12) or the baseline task (t[33] = 0.72, P = 0.48). However, the leftTLE subgroup were slower compared with controls for thesemantic task (t[32] = 5.93, P < 0.0001) and showed a trendtowards being slower for the baseline task (t[32] = 1.97, P = 0.06).Comparisons between the right TLE patients and the Controls(Matched) showed the same pattern: there were no differencesbetween the groups in terms of accuracy for the semantic task(t[34] = 1.18, P = 0.25) or the baseline task (t[34] = 0.25, P = 0.80),but there was a nonspecific slowing on both tasks (semantic:t[34] = 7.27, P < 0.0001; baseline: t[34] = 4.51, P < 0.0001). As withthe written word tasks, comparisons within the left and rightTLE subgroups for the picture trials revealed no significant differ-ences in terms of accuracy (semantic: t[31] = 0.47, P = 0.64; base-line: t[31] = 0.58, P = 0.57) or reaction time (semantic: t[30] = 0.79,P = 0.44; baseline: t[30] = 1.50, P = 0.14).

Importantly, the lack of differences in the performance ofthe left versus right TLE group could not be accounted for bythe time since surgery. Overall, the right TLE group had a longerperiod of recovery since surgery compared with the left TLEgroup (right TLE group: 7.5 years vs. left TLE group: 4.6 years;Table 1). To explore the potential impact of number of years

since surgery on semantic performance, we ran ANCOVAs onthe patient data including this variable as a covariate. Therewas no change in the direction of the results reported above forthe in-scanner behavioral performance. Both left and right TLEgroups performed the picture and written word semantic taskswith equal behavioral performance, both in terms of accuracyand correct response time. There was no effect of the covariateon performance. Full results for this analysis can be found inSupplementary Table S1.

Comparisons between the 2 presentation speeds confirmedthat, as intended, speeding up control participants increasedthe difficulty of the semantic and baseline tasks, reflected inreduced accuracy in both the picture (semantic: t[18] = 3.92, P =0.001; baseline: t[18] = 4.99, P < 0.0001) and written word condi-tion (semantic: t[18] = 4.38, P < 0.0001; baseline: t[18] = 5.47, P <0.0001).

Whole Brain Analyses

Written Word TasksRegions involved in the written word tasks in the control parti-cipants and patients were identified using the whole brain con-trasts “Written Words (CCw + Names) > Scrambled Words.”Peak activations for the whole brain contrasts are listed inSupplementary Table S2. Figure 3a shows an exclusively leftlateralized network activated by control participants by thewritten word semantic task. For control participants at thematched speed (top, red) this included the left IFG and posteriormiddle temporal gyrus. At the faster speed, controls showed anincrease in the extent of activation across the left hemisphere(top, green), particularly in the left IFG, posterior middle tempo-ral gyrus and fusiform gyrus extending into the left vATL.Medial structures including the dorsomedial prefrontal cortex,orbitofrontal cortex and precuneus were also implicated. Forthe left TLE patients, written word semantic tasks activatedbroadly the same areas as control participants (although notextending into the left vATL because of the nature of the resec-tion). In addition to these left hemisphere structures, left TLEpatients also showed activation to written words in the rightIFG (pars triangularis and pars orbitalis). For right TLE patients,written word tasks activated a largely left-lateralized network,which was very similar to the pattern of response seen in con-trol participants.

Table 2 In-scanner behavioral results. Mean accuracy (%) and correct response times (ms) across the experimental conditions (standard devi-ation in parenthesis). Note for the TLE patients and Controls (Matched) condition, stimuli were displayed on screen for 5000ms, for theControls (Speeded) condition stimuli were displayed for 2500ms.

Written word Picture Scrambled written Scrambled picture

% CorrectLeft TLE 88 (7)* 78 (7) 91 (4) 93 (7)Right TLE 91 (6) 79 (6) 88 (7) 92 (7)Controls (Matched) 94 (5) 81 (6) 89 (6) 91 (10)Controls (Speeded) 83 (11)** 75 (9)** 80 (9)** 81 (13)**

Correct response time (ms)Left TLE 2387 (378)* 2427 (344)* 2111 (422)* 2098 (468)*Right TLE 2396 (340)* 2520 (323)* 2329 (269)* 2307 (313)*Controls (Matched) 1714 (303) 1811 (262) 1791 (286) 1833 (317)Controls (Speeded) 1367 (150) 1272 (112) 1336 (130) 1346 (130)

*Significant difference between the TLE group vs. Control (Matched)—independent t-test. **Significant difference between the 2 control conditions—paired t-test.



Picture TasksRegions involved in the picture tasks in control participantsand patients were identified using the whole brain contrast“Picture (CCp + Faces) > Scrambled Pictures.” Peak activationsfor the whole brain contrasts are listed in SupplementaryTable S3. Figure 3b shows a widespread bilateral network ofregions activated. For control participants at the matched speed(top, red) this included the length of the temporal lobes bilater-ally, extending the length of the fusiform and inferior temporalgyri into the vATLs. At the slower speed, activation in the leftIFG, bilateral posterior middle temporal gyrus and precuneuswere also found. At the faster speed (green), control partici-pants showed a greater extent of activation in all these areas,particularly in the precuneus and left IFG. Additional activationwas also observed in the right IFG. For the left TLE patients,visual semantic tasks activated similar areas to control partici-pants, with the notable exception of left IFG; the right IFG wasalso active, however. For the right TLE patients, again similaractivation to the control participants was shown, and for thesepatients activation was found in the left but not right IFG.

ROI Analysis

Whole brain results illustrate that the overall pattern of activa-tion between postsurgical TLE patients and controls are broadlysimilar. Next, we tested whether TLE patients were over-activating or under-activating particular areas of interest incomparison to controls. Over-activation of a region may indi-cate a compensatory mechanism whereas under-activationmay indicate a negative impact.

Figure 4 shows the a priori ROI results for all participants.For the IFG ROI analysis, the analyses of interest were between-subjects ANOVAs comparing activation changes across thehemispheres (left, right) between the 2 TLE groups and Controls(Matched). For the ATL region, given the lack of data for the TLEsubgroups in one hemisphere, the analysis of interest wasindependent t tests comparing activation changes between the2 TLE subgroups and Controls (Matched) in the remaininghemisphere. Finally, paired t-tests comparing activationchanges as a function of presentation speed in controls. Bycomparing the pattern of activation between the 2 speeds incontrols, we are able to address the secondary question of this

study, which was to explore whether patients use a separateset of regions in order to carry out semantic memory tasks aftersurgery or whether they upregulate activity in the same regionsengaged by healthy individuals during more challengingsemantic processing.

Written Word TasksDuring written word processing, in all 3 left hemisphere ROIs(Fig. 4) there was a significant upregulation of activation in theControls (Speeded) condition compared with the Controls(Matched) condition (paired t-tests = vATL: t[19] = 4.27, P <0.0001; pars triangularis: t[19] = 3.07, P = 0.006; pars orbitalis:t[19] = 2.62, P = 0.017), indicating that controls increased activ-ity across the semantic network when task demands increased.

Compared with controls, TLE patients showed increasedactivation in a number of these areas. The right TLE groupshowed greater activation than controls in left IFG but not theleft ATL. In direct contrast to this, the left TLE group showedsignificant upregulation of activation compared with Controls(Matched) only in the right hemisphere. Specifically, in the IFG(pars orbitalis) a 2-way Group*Hemisphere ANOVA showed asignificant Group*Hemisphere interaction (F[2, 51] = 8.27, P =0.001). This was driven by an upregulation of activation in theleft IFG in the right TLE group compared with Controls(Matched; t[35] = 2.96, P = 0.006). In contrast to this, the left TLEgroup showed an upregulation in the right IFG compared withControls (Matched; t[35] = 2.34, P = 0.025) and the right TLEgroup (t[35] = 2.38, P = 0.023). In the ATL ROI there was no corre-sponding upregulation of activation in the left ATL in the rightTLE group compared with Controls (t[35] = 0.27, P = 0.79). Therewas a significant increase in activation in the right ATL in theleft TLE group (t[35] = 2.04, P = 0.049).

Picture Tasks

In contrast to the group differences shown in the written wordcondition, there were very few group differences in the picturecondition (Fig. 4). Comparisons between the 2 speeds in controlsrevealed no changes in activation in any of the ROIs, in either theleft or right hemisphere during picture trials. However, a 2-wayGroup*Hemisphere ANOVA performed on the IFG (pars triangu-laris) data showed a significant Group*Hemisphere interaction

Figure 3. Whole brain results for the 2 semantic tasks. (A) Whole brain maps for the written word > scrambled word contrast for the 2 control conditions (Top:

Matched—red, Speeded—green) and the 2 TLE patient groups (Bottom: left TLE—blue, right TLE—red). Resection areas for the 2 patients groups are shown in cyan;

the resection area for the right TLE group is not displayed on the ventral view to illustrate the right vATL activation for the left TLE group. (B) Whole brain maps for

the picture > scrambled picture contrast.

8 | Cerebral Cortex


Figure 4. A priori ROI results for the picture and written word semantic tasks. Contrast estimates for each task are shown over the relevant baseline condition for the

left and right TLE patients and the 2 control participant conditions (Matched vs. Speeded). Error bars denote standard error. vATL coordinates taken from Binney et al.

(2010), IFG (triangularis) coordinates taken from Noonan et al. (2013), IFG (orbitalis) coordinates taken from Badre et al. (2005)—ROI positions are shown in the top

right panel. “X” indicates where data are unavailable due to surgical resection. Full line = significant differences between the left or right TLE group and the Controls

(Matched) condition, dashed line = significant differences between the 2 control conditions (Matched vs. Speeded).



(F[2, 51] = 4.2, P = 0.02).This was driven by an upregulation of acti-vation in the right IFG in the left TLE group compared withControls (Matched; t[35] = 2.73, P = 0.01). The right TLE groupshowed no significant differences to the Controls (Matched) con-dition in any ROI.

DiscussionThis study explored neural reorganization of semantic functionin patients with unilateral ATL resection for TLE. The semanticrepresentation system is bilaterally organized, and is robust tounilateral damage/perturbation; however, the neural underpin-nings of this mechanism have yet to be fully elucidated. Oneview holds that residual semantic performance is maintainedvia upregulation of activation in the contralateral ATL(Schapiro et al. 2013; Binney and Lambon Ralph 2015; Jung andLambon Ralph 2016). An alternative, but not mutually exclu-sive, suggestion is that task performance is maintained viaupregulation of cognitive/semantic control regions (Backeset al. 2005; Maccotta et al. 2007; Wong et al. 2009; Bonelli et al.2012; Sidhu et al. 2013). Here we investigated the maintenanceof semantic performance for both word and picture stimuli in agroup of postsurgical left and right TLE patients, using an imag-ing protocol that improves signal in the vATLs (Visser,Embleton et al. 2010; Halai et al. 2014). The principal findingwas of dynamic changes in the semantic network after unilat-eral resection—supporting both types of proposed resiliencemechanisms. In particular, differential activation betweenpatients and control participants were shown in the IFG andvATLs. These changes were not uniform; rather the degree ofupregulation changed depending on the task being undertaken.Semantic tasks presented as pictures elicited bilateral activa-tion in both the control participants and TLE patients. The dis-tribution of activation was very similar, with the only groupdifference being stronger right IFG activation in the left TLEpatients compared with controls. These minimal group differ-ences in the fMRI data matched the pattern of results in thebehavioral data, which showed no differences in accuracyacross the TLE patients and controls. In contrast, during writtenword semantic tasks, activation was strongly lateralized to theleft hemisphere in both controls and patients (a pattern thathas been observed in large-scale meta-analyses of the semanticfMRI literature in healthy participants: Rice, Lambon Ralphet al. 2015). More striking group differences between thepatients and controls were revealed in this condition. The rightTLE group upregulated the left IFG compared with controls;whereas, the left TLE group upregulated the homologous rightIFG and right vATL. This differential fMRI activation profile inthe left TLE group reflected the behavioral findings, whichshowed less accurate responses in the left TLE group in writtenword tasks. The results suggest, therefore, that 1) the patientsupregulate both the representational (contralateral ATL) andexecutive control (IFG) systems in a manner that seems toreflect intrinsic mechanisms which make the healthy systemresilient to increased demand (cf. the speeded condition inhealthy individuals), but 2) these effects interact with both taskmodality and the side of resection.

The semantic representation system is intrinsically robustto unilateral damage compared with bilateral damage (Kluverand Bucy 1937; 1938; Terzian and Ore 1955; Wilkins andMoscovitch 1978; Lambon Ralph et al. 2010, 2012). Unilateralresection results in a mild semantic impairment which is mostreliably shown through slower response times on semantictasks (Wilkins and Moscovitch 1978; Lambon Ralph et al. 2012;

Rice et al. 2018). The adaptive properties of the semantic sys-tem after unilateral resection/perturbation have been ascribedto the upregulation of activation in the remaining contralateralATL which helps to maintain the robustness of the system(Schapiro et al. 2013; Binney and Lambon Ralph 2015; Jung andLambon Ralph 2016). This mechanism of contralateral upregu-lation has also been proposed as a mechanism within themedial temporal lobes for the maintenance of episodic memoryfunction (Bonelli et al. 2012, 2013; Sidhu et al. 2013). Here, weshowed some changes in the remaining ATL in TLE patients,although these changes interacted with the task demands.Semantic tasks presented as pictures elicited a strongly bilat-eral representation in the left and right vATLs in healthy parti-cipants, in keeping with previous observations of tasksrequiring a response to pictures (Rice, Lambon Ralph et al.2015). Under increased task demands, no upregulation wasseen in the left or right vATLs in control participants. Similarly,after resection of either the left or right ATL, no upregulation ofactivation in the remaining ATL was found in for patients. Thisimplies that information that is represented bilaterally is morerobust to unilateral damage, such that the contralateral ATLmay retain the conceptual representations or sufficient noise-free fidelity needed to complete the task without the need forupregulation (Schapiro et al. 2013).

In contrast, written word semantic tasks elicited a stronglyleft lateralized network in healthy participants, which again isin line with previous neuroimaging results (Rice, Lambon Ralphet al. 2015). Under increased task demands in control partici-pants, there was strong upregulation in the left vATL, but notthe right. The right TLE group also activated a strongly leftlateralized network during written word processing. In con-trast, the left TLE patients activated the right vATL more thancontrol participants (directly mirroring the 2 previous combinedrTMS-fMRI explorations in healthy participants, which also uti-lized written word materials) (Binney and Lambon Ralph 2015;Jung and Lambon Ralph 2016), indicating reorganization of ver-bal semantic processing in these patients to make use of thecontralateral ATL. These patients were significantly impaired incompletion of the verbal task, implying that tasks that are rela-tively lateralized to one hemisphere are more vulnerable tounilateral damage/resection of the dominant hemisphere (seealso, Schapiro et al. 2013). Although representations stored inthe contralateral ATL do appear to be recruited in thesepatients, in this case they were not sufficient to maintain nor-mal performance.

In addition to an intact ATL representational store, success-ful semantic cognition requires executive control processesthat serve to shape this information in a task appropriate man-ner (Jefferies and Lambon Ralph 2006; Jefferies 2013; Rogerset al. 2015). After degradation of the semantic store (throughunilateral ATL resection or perturbation), upregulation in con-trolled retrieval mechanisms in the bilateral IFG may becomenecessary in order to maintain normal task performance (Koyluet al. 2006; Bonelli et al. 2012; Sidhu et al. 2013). Here, reliableactivation changes were revealed in 2 separate areas of the IFG.One cluster was located in pars orbitalis (BA47) and the secondwas located in pars triangularis (BA45). As in the vATL, activa-tion changes in the IFG interacted with task modality. In thepicture-based semantic task there were few group differences.More striking group differences were shown in the writtenword task; specifically, the right TLE group showed an upregu-lation of activation in the left IFG (pars orbitalis) compared withcontrol participants. In contrast, left TLE patients increasedactivation in the right IFG, homologous to regions upregulated

10 | Cerebral Cortex


by right TLE patients. The differential IFG activation profile inTLE patients versus control participants accords with previousfindings in postsurgical TLE patients (Backes et al. 2005; Koyluet al. 2006; Bonelli et al. 2012) and suggests that upregulation ofsemantic control regions may be important for supportingsemantic processing in patients with ATL resection for TLE. Itmay be worth noting that the specific upregulated PFC regionswere the pars orbitalis. Previous studies of healthy semanticfunction (Badre et al. 2005) found that these regions were cru-cial when the target semantic information was inherently weakand presumably required a form of “amplification.” This is con-sistent with the possibility that the patients’ long-term ATLdysfunction and later resection weakens their semantic repre-sentations which can be boosted via interaction with theseventral PFC semantic control regions.

A related question is whether the areas showing dynamicchanges are specific to the patients or whether such changesare part of the normal mechanisms that are used in the healthysystem. In other words, recovery may involve reoptimization ofexisting resources whereby contralateral regions that werealready somewhat involved in supporting the affected func-tions upregulate their contribution to compensate for the dam-age. Evidence for this type of upregulation in the contralateralhemisphere has been found in the language/semantic domainafter stroke (Leff et al. 2002; Warren et al. 2009) and for long-term recovery of episodic memory function after resection forTLE (Bonelli et al. 2012, 2013; Sidhu et al. 2016). In neurologicallyintact participants, this same mechanism has been demon-strated for semantic aspects of language (Sharp et al. 2004).Brownsett et al. (2014) went one step further and showed thatregions in the frontal lobes that were upregulated during lan-guage processing in poststroke aphasia overlapped withregions that healthy controls upregulated when listening todegraded speech. This supports the general view that patientsincrease engagement of demand-related prefrontal regions tocompensate for damage elsewhere. Future research shouldexplore the interaction between the frontal and temporal lobesemantic systems in maintaining normal semantic perfor-mance, not only in patient populations but in the healthysemantic system.

Finally, we note that from the current study it is not possibleto determine whether the differences in activation between theTLE patients and control participants were caused by the sur-gery itself or whether activation patterns in TLE patients wereaffected prior to surgery as a result of long-standing epilepsy(or a combination of the 2). The purpose of the current studywas to establish changes in the function of the semantic net-work in a group of chronic postsurgery TLE patients. The pur-pose of the study was not to explore the root causes behindany patient versus control group differences, for which presur-gical data in the patients is required. Follow-up studies can usethis paradigm to explore semantic abilities both before andafter TLE resection with respect to both the patients’ semanticperformance and its neural bases. These types of studies wouldallow us to explore whether there are semantic deficits presur-gery, and whether or not these are exacerbated by the resectionsurgery. The effect of surgery on different semantic tasks (e.g.,pictures vs. written words) should also be a future avenue forresearch.

Supplementary MaterialSupplementary material is available at Cerebral Cortex online.

FundingG.E.R. was supported by a PhD studentship from EPSRC and aPresident’s Doctoral Scholarship from the University ofManchester. This research was supported by Medical ResearchCouncil programme grants (MR/J004146/1 and MR/R023883/1)and ERC grant (GAP: 670428-BRAIN2MIND_NEUROCOMP) toM.A.L.R. P.H. is supported by The University of EdinburghCentre for Cognitive Ageing and Cognitive Epidemiology, partof the cross council Lifelong Health and Wellbeing Initiative(MR/K026992/1). Funding from the Biotechnology and BiologicalSciences Research Council (BBSRC) and Medical ResearchCouncil (MRC) is gratefully acknowledged.

NotesWe thank the all the TLE patients and their families for gener-ously giving up their time to participate in this study. We alsothank Danielle Bowden and Lucy Palmer for assistance inpatient recruitment. Conflict of Interest: None declared.

ReferencesAlpherts WCJ, Vermeulen J, van Rijen PC, da Silva FHL, van

Veelen CWM, Dutch Collaborative Epilepsy S. 2006. Verbalmemory decline after temporal epilepsy surgery? A 6-yearmultiple assessments follow-up study. Neurology. 67(4):626–631.

Backes WH, Deblaere K, Vonck K, Kessels AG, Boon P, HofmanP, Wilmink JT, Vingerhoets G, Boon PA, Achten R, et al. 2005.Language activation distributions revealed by fMRI in post-operative epilepsy patients: differences between left- andright-sided resections. Epilepsy Res. 66(1–3):1–12.

Badre D, D’Esposito M. 2009. Is the rostro-caudal axis of thefrontal lobe hierarchical? Nat Rev Neurosci. 10(9):659–669.

Badre D, Poldrack RA, Pare-Blagoev EJ, Insler RZ, Wagner AD.2005. Dissociable controlled retrieval and generalized selec-tion mechanisms in ventrolateral prefrontal cortex. Neuron.47(6):907–918.

Badre D, Wagner AD. 2002. Semantic retrieval, mnemonic control,and prefrontal cortex. Behav Cogn Neurosci Rev. 1:206–218.

Billingsley RL, McAndrews MP, Crawley AP, Mikulis DJ. 2001.Functional MRI of phonological and semantic processing intemporal lobe epilepsy. Brain. 124:1218–1227.

Binney RJ, Embleton KV, Jefferies E, Parker GJ, Lambon Ralph MA.2010. The ventral and inferolateral aspects of the anteriortemporal lobe are crucial in semantic memory: evidence froma novel direct comparison of distortion-corrected fMRI, rTMS,and semantic dementia. Cereb Cortex. 20(11):2728–2738.

Binney RJ, Lambon Ralph MA. 2015. Using a combination offMRI and anterior temporal lobe rTMS to measure intrinsicand induced activation changes across the semantic cogni-tion network. Neuropsychologia. 76:170–181.

Bonelli SB, Thompson PJ, Yogarajah M, Powell RHW, SamsonRS, McEvoy AW, Symms MR, Koepp MJ, Duncan JS. 2013.Memory reorganization following anterior temporal loberesection: a longitudinal functional MRI study. Brain. 136:1889–1900.

Bonelli SB, Thompson PJ, Yogarajah M, Vollmar C, Powell RHW,Symms MR, McEvoy AW, Micallef C, Koepp MJ, Duncan JS.2012. Imaging language networks before and after anteriortemporal lobe resection: results of a longitudinal fMRI study.Epilepsia. 53(4):639–650.



Bozeat S, Lambon Ralph MA, Patterson K, Garrard P, Hodges JR.2000. Non-verbal semantic impairment in semantic demen-tia. Neuropsychologia. 38(9):1207–1215.

Brown S, Schafer EA. 1888. An investigation into the functionsof the occipital and temporal lobes of the monkey’s brain.Philos Trans R Soc Lond B Biol Sci. 179:303–327.

Brownsett SLE, Warren JE, Geranmayeh F, Woodhead Z, LeechR, Wise RJS. 2014. Cognitive control and its impact on recov-ery from aphasic stroke. Brain. 137:242–254.

Devlin JT, Russell RP, Davis MH, Price CJ, Wilson J, Moss HE,Matthews PM, Tyler LK. 2000. Susceptibility-induced loss ofsignal: comparing PET and fMRI on a semantic task.Neuroimage. 11(6):589–600.

Duncan J. 2010. The multiple-demand (MD) system of the pri-mate brain: mental programs for intelligent behaviour.Trends Cogn Sci. 14(4):172–179.

Embleton KV, Haroon HA, Morris DM, Lambon Ralph MA, ParkerGJM. 2010. Distortion correction for diffusion-weighted MRItractography and fMRI in the temporal lobes. Hum BrainMapp. 31(10):1570–1587.

Gainotti G. 2007. Different patterns of famous people recogni-tion disorders in patients with right and left anterior tempo-ral lesions: a systematic review. Neuropsychologia. 45(8):1591–1607.

Gainotti G. 2012. The format of conceptual representations dis-rupted in semantic dementia: a position paper. Cortex. 48(5):521–529.

Geranmayeh F, Brownsett SLE, Wise RJS. 2014. Task-inducedbrain activity in aphasic stroke patients: what is drivingrecovery? Brain. 137:2632–2648.

Halai AD, Welbourne SR, Embleton KV, Parkes LM. 2014. A com-parison of dual gradient-echo and spin-echo fMRI of theinferior temporal lobe. Hum Brain Mapp. 35:4118–4128.

Jefferies E. 2013. The neural basis of semantic cognition: con-verging evidence from neuropsychology, neuroimaging andTMS. Cortex. 49(3):611–625.

Jefferies E, Lambon Ralph MA. 2006. Semantic impairment instroke aphasia versus semantic dementia: a case-seriescomparison. Brain. 129(Pt 8):2132–2147.

Jung J, Lambon Ralph MA. 2016. Mapping the dynamic networkinteractions underpinning cognition: a cTBS-fMRI study ofthe flexible adaptive neural system for semantics. CerebCortex. 26:3580–3590.

Kluver H, Bucy PC. 1937. “Psychic blindness” and other symp-toms following bilateral temporal lobectomy in rhesus mon-keys. Am J Physiol. 119:352.

Kluver H, Bucy PC. 1938. An analysis of certain effects of bilat-eral temporal lobectomy in the rhesus monkey, with specialreference to “Psychic Blindness”. J Psychol. 5(1):33–54.

Koylu B, Trinka E, Ischebeck A, Visani P, Trieb T, Kremser C,Bartha L, Schocke M, Benke T. 2006. Neural correlates of ver-bal semantic memory in patients with temporal lobe epi-lepsy. Epilepsy Res. 72(2–3):178–191.

Lambon Ralph MA, Cipolotti L, Manes F, Patterson K. 2010.Taking both sides: do unilateral anterior temporal lobelesions disrupt semantic memory? Brain. 133(11):3243–3255.

Lambon Ralph MA, Ehsan S, Baker GA, Rogers TT. 2012.Semantic memory is impaired in patients with unilateralanterior temporal lobe resection for temporal lobe epilepsy.Brain. 135(1):242–258.

Lambon Ralph MA, Jefferies E, Patterson K, Rogers TT. 2017. Theneural and computational bases of semantic cognition. NatRev Neurosci. 18(1):42–55.

Leff A, Crinion J, Scott S, Turkheimer F, Howard D, Wise R. 2002.A physiological change in the homotopic cortex followingleft posterior temporal lobe infarction. Ann Neurol. 51(5):553–558.

Maccotta L, Buckner RL, Gilliam FG, Ojemann JG. 2007.Changing frontal contributions to memory before and aftermedial temporal lobectomy. Cereb Cortex. 17(2):443–456.

Mion M, Patterson K, Acosta-Cabronero J, Pengas G, Izquierdo-Garcia D, Hong YT, Fryer TD, Williams GB, Hodges JR, NestorPJ. 2010. What the left and right anterior fusiform gyri tell usabout semantic memory. Brain. 133(11):3256–3268.

Noonan KA, Jefferies E, Visser M, Lambon Ralph MA. 2013.Going beyond inferior prefrontal involvement in semanticcontrol: evidence for the additional contribution of dorsalangular gyrus and posterior middle temporal cortex. J CognNeurosci. 25(11):1824–1850.

Noppeney U, Price CJ, Duncan JS, Koepp MJ. 2005. Reading skillsafter left anterior temporal lobe resection: an fMRI study.Brain. 128:1377–1385.

Pobric G, Jefferies E, Lambon Ralph MA. 2010. Amodal semanticrepresentations depend on both anterior temporal lobes:evidence from repetitive transcranial magnetic stimulation.Neuropsychologia. 48(5):1336–1342.

Rice GE, Caswell H, Moore P, Hoffman P, Lambon Ralph MA.2018. The roles of left versus right anterior temporal lobes insemantic memory: a neuropsychological comparison ofpostsurgical temporal lobe epilepsy patients. Cereb Cortex.28(4):1487–1501.

Rice GE, Hoffman P, Lambon Ralph MA. 2015a. Graded speciali-zation within and between the anterior temporal lobes. AnnN Y Acad Sci. 1359:84–97.

Rice GE, Lambon Ralph MA, Hoffman P. 2015b. The roles of theleft vs. right anterior temporal lobes in conceptual knowl-edge: an ALE meta-analysis of 97 functional neuroimagingstudies. Cereb Cortex. 25(11):4374–4391.

Rogers TT, Patterson K, Jefferies E, Lambon Ralph MA. 2015.Disorders of representation and control in semantic cogni-tion: effects of familiarity, typicality and specificity.Neuropsychologia. 76:220–239.

Rosazza C, Ghielmetti F, Minati L, Vitali P, Giovagnoli AR, DeleoF, Didato G, Parente A, Marras C, Bruzzone MG, et al. 2013.Preoperative language lateralization in temporal lobe epi-lepsy (TLE) predicts peri-ictal, pre- and post-operative lan-guage performance: an fMRI study. NeuroImage Clin. 3:73–83.

Schapiro AC, McClelland JL, Welbourne SR, Rogers TT, LambonRalph MA. 2013. Why bilateral damage is worse than unilat-eral damage to the brain. J Cogn Neurosci. 25(12):2107–2123.

Seghier ML, Ramlackhansingh A, Crinion J, Leff AP, Price CJ.2008. Lesion identification using unified segmentation-normalisation models and fuzzy clustering. Neuroimage.41(4):1253–1266.

Sharp DJ, Scott SK, Wise RJS. 2004. Retrieving meaning aftertemporal lobe infarction: the role of the basal language area.Ann Neurol. 56(6):836–846.

Shimotake A, Matsumoto R, Ueno T, Kunieda T, Saito S,Hoffman P, Kikuchi T, Fukuyama H, Miyamoto S, TakahashiR, et al. 2014. Direct exploration of the role of the ventralanterior temporal lobe in semantic memory: cortical stimu-lation and local field potential evidence from subdural gridelectrodes. Cereb Cortex. 25(10):3802–3817.

Sidhu MK, Stretton J, Winston GP, Bonelli S, Centeno M,Vollmar C, Symms M, Thompson PJ, Koepp MJ, Duncan JS.2013. A functional magnetic resonance imaging study

12 | Cerebral Cortex


mapping the episodic memory encoding network in tempo-ral lobe epilepsy. Brain. 136:1868–1888.

Sidhu MK, Stretton J, Winston GP, McEvoy AW, Symms M,Thompson PJ, Koepp MJ, Duncan JS. 2016. Memory networkplasticity after temporal lobe resection: a longitudinal func-tional imaging study. Brain. 139:415–430.

Sidhu MK, Stretton J, Winston GP, Symms M, Thompson PJ,Koepp MJ, Duncan JS. 2015. Factors affecting reorganisationof memory encoding networks in temporal lobe epilepsy.Epilepsy Res. 110:1–9.

Terzian H, Ore GD. 1955. Syndrome of Kluver and Bucy—repro-duced in man by bilateral removal of the temporal lobes.Neurology. 5(6):373–380.

Thompson-Schill SL, D’Esposito M, Aguirre GK, Farah MJ. 1997.Role of left inferior prefrontal cortex in retrieval of semanticknowledge: a reevaluation. Proc Natl Acad Sci U S A. 94(26):14792–14797.

Tivarus ME, Starling SJ, Newport EL, Langfitt JT. 2012.Homotopic language reorganization in the right hemisphereafter early left hemisphere injury. Brain Lang. 123(1):1–10.

Visser M, Embleton KV, Jefferies E, Parker GJ, Lambon RalphMA. 2010a. The inferior, anterior temporal lobes and seman-tic memory clarified: novel evidence from distortion-corrected fMRI. Neuropsychologia. 48(6):1689–1696.

Visser M, Jefferies E, Embleton KV, Lambon Ralph MA. 2012. Boththe middle temporal gyrus and the ventral anterior temporalarea are crucial for multimodal semantic processing:distortion-corrected fMRI evidence for a double gradient of

information convergence in the temporal lobes. J CognNeurosci. 24(8):1766–1778.

Visser M, Jefferies E, Lambon Ralph MA. 2010b. Semantic pro-cessing in the anterior temporal lobes: a meta-analysis ofthe functional neuroimaging literature. J Cogn Neurosci.22(6):1083–1094.

Warren JE, Crinion JT, Lambon Ralph MA, Wise RJS. 2009.Anterior temporal lobe connectivity correlates with func-tional outcome after aphasic stroke. Brain. 132:3428–3442.

Wiebe S, Blume WT, Girvin JP, Eliasziw M, EffectivenessEfficiency Surgery T. 2001. A randomized, controlled trial ofsurgery for temporal-lobe epilepsy. N Engl J Med. 345(5):311–318.

Wilkins A, Moscovitch M. 1978. Selective impairment of seman-tic memory after temporal lobectomy. Neuropsychologia.16(1):73–79.

Willment KC, Golby A. 2013. Hemispheric lateralization inter-rupted: material-specific memory deficits in temporal lobeepilepsy. Front Hum Neurosci. 7:8.

Wong SWH, Jong L, Bandur D, Bihari F, Yen YF, Takahashi AM,Lee DH, Steven DA, Parrent AG, Pigott SE, et al. 2009. Corticalreorganization following anterior temporal lobectomy inpatients with temporal lobe epilepsy. Neurology. 73(7):518–525.

Zhao Y, Song LP, Ding JH, Lin N, Wang Q, Du XX, Sun R, Han Z.2017. Left anterior temporal lobe and bilateral anterior cin-gulate cortex are semantic hub regions: evidence frombehavior-nodal degree mapping in brain-damaged patients.J Neurosci. 37(1):141–151.



Documents

Edinburgh Research Explorer · changes in the left and right ATLs, there was increased effec-tive connectivity from the right ATL to the left ATL (Jung and Lambon Ralph 2016). In